Gaussian Mixture Model based Deep Learning-based Simple Online Real-Time Tracking for Video Object Detection and Tracking

Nowadays, the increasing demand for intelligent video analysis has driven the need for accurate and efficient video object detection and tracking systems. However, the existing You Only Look Once (YOLO) model have limitations in handling large-scale video data and modeling complex object interactions. Hence, this research proposes Deep Neural Network with Gaussian Mixture Model based Deep Learning-based Simple Online Real-Time Tracking (GMM-DeepSORT) to handle the large-scale video data. The proposed GMM-DeepSORT detect and track objects across frames for accurate video object detection and tracking. Initially, the input video data is collected from Large-scale Single Object Tracking (LaSOT) dataset which consists of video sequences by providing both visual and lingual annotation for each sequence. Then, GMM is employed to eliminate the background from videos and detect all the moving objects in input video frames. Finally, the DeepSORT is introduced to track the objects by predicting object motion and associating object across video frames. From the results, the proposed GMM-DeepSORT established better results in terms of accuracy (97.92%) and processing time for tracking (0.31s) when compared to Deeply Supervised Object Detector (DSOD) respectively.